Research is a
25 volumes · 4 tracks
What we learned building shit. All of it, free.
RSSThe index
25 volumes
-
PH—025 Distribution Engineering Vol 25 Distribution EngineeringKeywords Outlived the Click
Keyword research was built to win a visit, and the visit is now reduced, uncounted or mislabelled. We still do the research, as a study of what people want and what they call it. That changes which terms we chase, how far we trust a referral report, and how we write the one press release.
-
PH—024 Distribution Engineering Vol 24 Distribution EngineeringThe Trusted Network Is the First Instrument, Not a Shortcut
Founders lean on their network because it answers fast and tells the truth about what is confusing. We use it the same way, and we record every warm result as a ceiling. The network tells us what is true about an offer. Cold traffic tells us whether it travels.
-
PH—023 Distribution Engineering Vol 23 Distribution EngineeringAI Multiplied the Experiments, Not the Answers
Agents made test variants nearly free. Traffic, time and statistical power cost what they always did, so teams start far more tests than they finish. At small sample sizes the right use of an agent is fewer, larger bets and a decision rule written before the data arrives.
- Vol 22 Distribution Engineering
How to Market an AI Tool
You built an AI tool and now someone has to use it. This is the order we would do it in: write an offer that can fail, pick fifty people by hand, test the page before the product, and treat launch day and the press release as jobs with a purpose. It also covers when to stop and change the offer.
- Vol 21 Distribution Engineering
Distribution Engineering Is a Discipline
Distribution engineering is specifying, building and instrumenting a product's route to people with the same rigour as the product. It has three parts: an offer written as a claim that can fail, a test with a threshold set in advance, and a measured handoff into operations. Building got cheap. The route did not.
- Vol 20 Commercialization Patterns
The Anti-Consulting Manifesto.
Twenty volumes. Thirty installs. Seven verticals. Here is what we have concluded. Enterprise AI is a software problem dressed as a consulting problem. The right answer is software. And the industry that got it wrong has five years to catch up — or watch a new kind of studio eat its lunch.
- Vol 19 Commercialization Patterns
AI Vendor Risk · A Buyer’s Checklist.
Most AI vendors sell you a dependency. Here is the checklist every buyer should run before signing. Twelve questions that separate software you can keep from software that leaves when the vendor does.
- Vol 18 Commercialization Patterns
The Swappable Base.
The foundation model is a commodity. We treat it like one. Here is the architecture that lets the base swap out — Opus to Sonnet, GPT-4 to Gemini, open weights fine-tune — without anything above it breaking.
- Vol 17 Commercialization Patterns
Invisible Design, Revisited.
The best interface is the one the user never thinks about. A year of install work has sharpened that claim. Here are the four things we now delete by default, and why the best AI products refuse to show off.
- Vol 16 Industry Ontologies
Evaluating AI in Regulated Environments.
HIPAA, SOC 2, GDPR, FDA 21 CFR Part 11, ISO 27001. The compliance filter does not slow the install. It shapes what the recommender is even allowed to propose. Here is how.
- Vol 15 Install Field Reports
Install Field Report · Home Inspection RAG.
A RAG system over regional inspection reports. The retrieval side was easy. The citation side was not. Sourced answers are the product. Here is what we shipped, where retrieval failed, and how the ontology caught it.
- Vol 14 Commercialization Patterns
Retraining Without Regression.
The corpus updates quarterly. That is a promise, not a risk. Here is how we retrain without breaking what already shipped, and why regression is the term we wake up thinking about.
- Vol 13 Industry Ontologies
Seven Verticals, Seven Ontologies.
Every vertical has its own regulatory reality, economic reality, and operating vocabulary. A generalist model approximates all three. An ontology codes them. This is the table, one filter at a time, across the seven verticals we currently ship.
- Vol 12 Install Field Reports
Install Field Report · Concierge.
A hotel concierge AI answers guests, escalates to staff, and never invents a name. The design constraint is stronger than the model choice. Here is what we installed, what it refuses, and why the refusals are the product.
- Vol 11 Commercialization Patterns
The Build / Hire / Automate Decision.
Build, hire, or automate. The three lanes are not equivalent. Each has a cost curve, a timeline, and a failure mode. Here is how Cost Analysis models the trade on our installs, and the one thing every in-house spreadsheet leaves out.
- Vol 10 Commercialization Patterns
Context Engineering for 2026.
Prompt engineering was never the practice. Context engineering is. How the 2026 Corpus, a live readiness score, and a per-industry ontology compose into context the agent actually uses. And what it costs, in tokens and in latency, to get this right.
- Vol 09 Install Field Reports
Install Field Report · Clarity SI.
Property intelligence is a dashboard problem disguised as an AI problem. Clarity SI proved that the right aggregation beats the fancier model most of the time. One screen. The number leadership wanted.
- Vol 08 Commercialization Patterns
Building Agents You Can Inspect.
Agents that work but cannot be inspected are not production software. They are spells. Here is what inspectable means on our install, and why it is the difference between an agent your team operates and one they tolerate.
- Vol 07 Commercialization Patterns
Why the Base Model Isn’t the Moat.
Every competitor can buy the same foundation model tomorrow. The moat is what sits on top. This is the four-layer architecture we install, and why it compounds in calendar time rather than engineering time.
- Vol 06 Install Field Reports
Install Field Report · Roadrunner Septic.
Trade services is not fintech. The install proved that the ontology filter works as hard on a septic booking flow as it does on a compliance workflow. Different surface. Same spine. Here is what held, what broke, and what the install taught the corpus.
- Vol 05 Commercialization Patterns
Agentic RAG in Production.
Retrieval and reasoning are two different problems. Most teams solve the first and pretend it solved the second. Here is where retrieval should stop. Here is where the agent should start. Drawn from the Home Inspection RAG install, and five others that went the same way.
- Vol 04 Industry Ontologies
One Ontology per Industry.
A healthcare commercialization recommendation does not look like a PE portfolio one. The model knows why. The reason is not in the base model. It is in the translator layer that sits above it. This is the shape of that layer, and why a generalist model cannot approximate it without it.
- Vol 03 Install Field Reports
Install Field Report · Port of San Antonio.
UX discovery and readiness audit ran in parallel on the Port of San Antonio install. What surfaced in the audit changed the UX work. What surfaced in the UX work changed the ontology. Here is what we found.
- Vol 02 Commercialization Patterns
What a Readiness Score Actually Measures.
The readiness score is the first number a Terrein install produces. It is not a grade. It measures the distance between an organization’s operating reality and its operating ambition. Here is what the number contains, how it moves, and why eighty-seven out of a hundred means something.
- Vol 01 Commercialization Patterns
The AI Pilot Is Over.
Enterprises aren't failing at AI. They're failing at AI delivery. After thirty installs in five months, the pattern is clear. The companies winning with AI have stopped piloting and started installing. This is a field note about what we saw, what we changed, and what we ship now.
From the archive
5 essays · 2025
- 9 min Architecture
AI Agentic RAG
Retrieval-augmented generation has graduated from a trick to an architecture. How agentic RAG lets systems reason about what they don't know — and when to go find it.
- 10 min Technical
Amazon Bedrock: A Practitioner's Read
What Bedrock gets right, what it still doesn't, and how we use it in production alongside Claude and OpenAI. Notes from shipping four systems on the platform.
- 8 min Design
Invisible Design, Visible Impact
How great UX builds trust. A close reading of the micro-decisions — defaults, copy, feedback latency — that separate products users return to from products they tolerate.
- 7 min Philosophy
Why AI Should Help, Not Just Hype
A critique of the industry's theatrical reflex. What it looks like to ship AI that actually compresses a user's time rather than expanding the vendor's narrative.
- 12 min Essay
The AI Assistant User Experience Blueprint
Trust is the product. A framework for designing assistants that earn user delegation without overpromising — covering affordances, failure states, and the ethics of proactivity.